Aidan Mercer

What did diligence and portfolio work look like before you brought AI into it?

Diligence was mostly reading. A data room for a lower-middle-market company might hold a few thousand files: financial statements, contracts, customer lists, HR records, a management presentation. I worked through them in a set order, took notes, and built a picture of the business one document at a time. The slow part was not the analysis, it was retrieval. If a question came up on Thursday about customer concentration, I would go hunting through folders I had last opened on Monday.

Portfolio work had the same shape. Twelve companies sent monthly reports in twelve formats, and I spent the first week of each month reconciling them by hand before I could see anything. The work was thorough, but a lot of my hours went to finding and formatting rather than to judgment.

What was the first thing you tried, and how did it go?

The first real test was loading one company's diligence documents into a tool I could ask questions in plain language. Instead of searching for a filename, I asked, "Which customers are more than ten percent of revenue, and what are their contract terms?" It pulled the answer and pointed me to the pages it came from. That last part mattered more than the speed. My early worry was that the model would sound confident and be wrong, so I only trusted it when it cited a source I could open and verify.

It got some things imprecise, usually where a number lived in a footnote or a spreadsheet tab. But the first afternoon convinced me. I got through that data room in a day instead of most of a week, and I read more of it, not less.

Walk me through how you actually use it in diligence now.

I treat it as a research assistant that never gets tired of the data room. Everything goes into a secured, approved environment, not a public chatbot. I start broad: summarize the business, list the top risks in these financials, reconcile the revenue figures across the model and the audited statements. Then I get specific and interrogate what it surfaces. Where are gross margins slipping and why. Which contracts have change-of-control clauses. What does churn look like by cohort.

Every answer has to trace back to a document, and I open the document. The tool builds the chronology and the first list of issues; I decide which ones are real and which ones change the price. It has not replaced a single judgment. It has moved my time from assembling the picture to arguing with it.

How does it change the portfolio side of the job?

Portfolio monitoring used to be a lag indicator. By the time I had reconciled twelve reports, the month was over. Now the KPIs land in one place and the model does the first read: it flags that a company's customer acquisition cost jumped, that cash runway is under six months, that a segment's margin moved against plan. I still call the CEO and ask what happened, because a flag is a question, not a conclusion. But I am asking it in week one instead of week three.

It also drafts the analysis I take into a board meeting, so I walk in with the anomalies already mapped and my own read on which ones matter. The analysis is deeper because I am not spending those hours on data cleanup.

Where don't you trust it, and where do you keep control?

Two hard lines. The first is confidentiality. I handle material nonpublic information: target financials, valuation models, limited-partner communications. None of that goes into a consumer tool that might train on it or leak it. It stays inside an approved, access-controlled environment with the same NDAs and security review as the rest of the deal, because exposing MNPI is a securities-law and fiduciary problem, not only an IT one.

The second is accountability. The model does not make the investment call and does not sign the memo. It can miss the thing that is not written down: the customer about to leave, the founder dynamic that will not survive a transaction. So I verify every number that matters against its source, and the recommendation is mine. If it is wrong, that is on me, not the tool, and I want it that way.

Was there a moment it clearly earned its place?

One deal, late in diligence, on a tight exclusivity clock. The seller dropped a large batch of supplementary contracts into the data room over a weekend, right before our investment committee. Reading all of them by Monday was not realistic by hand. I ran them through the tool and asked it to list every non-standard clause, every auto-renewal, and every customer with unusual termination rights, each cited to its page.

It surfaced a change-of-control provision in a major customer agreement that would have let that customer walk on the sale. That single clause changed how we structured the deal. I would like to say I would have found it anyway, and maybe I would have, at two in the morning. The honest version is that the tool got me there with hours to spare, and I had time to verify it and think.

What would you tell a partner who thinks this is hype?

That it is a research assistant, not an oracle, and the value shows up only if you keep reading. It does not think for you. It reads faster than you, it never loses a document, and it will point you to the page every time if you insist on citations. Used that way, it gives back the hours that went to retrieval and formatting, and you spend them on the parts of the job that need a partner: judgment, negotiation, the conversation with a management team.

I would also tell them the risk is real and specific. Put confidential deal data somewhere it can leak and you have a serious problem. Keep it in a controlled environment, verify what matters, and own the decision, and you get the upside without betting the firm's reputation on a model.